Ojas Shukla

You will only care about the model Everything you do is for the model > Build a startup, service of the model > Work a 9-5, employed by the model > Spearfish in the Arctic, believe it or not... sustaining the model This post is for the model. Nothing matters but the model.

ethereum is perfect architecture trapped inside a terrible product. I've been deploying contracts for the longest time: go to arbitrum because hype's there bridge tokens to base for another deploy hope this L(𝒩)Ⓡ chain still exists in 6 months the devs are there. the network effects are there. all the top chains barring a few are EVM. every builder knows the idea is right. the tech works. the product doesn't. mainnet's unusable, liquidity's split across a hundred chains. EVM already won. the product just hasn't shipped yet.

The Strait of Hormuz is just CowSwap for the global economy.

Using an LLM to write on X is the same as using one to trade You only get beta Alpha is the part of you that isn't in the training data

Proud to have the chance to help build some of the early product here @farbood's been biohacking for 25 years & @BenMaxR built Instagram's camera/AR org at Meta. Incredible team shipping a genuinely different health product - everything consolidated & backed by real research

@aryanXmahajan @BrianNorgard Coming from a real one!

@synthpolis @BrianNorgard Legend

Wrote SporeMesh with @BrianNorgard. I want to explain what this is for people seeing it for the first time. So Karpathy released autoresearch a couple days ago- really cool concept! It's a simple, but novel, script where an AI agent looks at its own training code, comes up with an idea to improve it, rewrites the code, trains for five minutes, checks if it actually got better, and does it again. ~100 experiments overnight, no human involved. Shopify's CEO (@tobi) already ran it and got a 19% improvement -- the agent-tuned small model beat a larger one tuned by hand. The limitation is it runs on a single GPU. One agent, one thread of exploration. Karpathy pointed this out himself -- the next step has to be massively collaborative, like SETI@home. Not one PhD student, a whole research community. That's SporeMesh Easiest way to think about it: LimeWire for autoresearch. You run a node on whatever GPU you have, even a consumer card. Your node runs that same loop - rewrite, train, evaluate - but shares what it learns with every other node on the network. The whole network gets smarter together. The thing that gets me excited about this: frontier research doesn't actually need the biggest GPU. It needs entropy-- diversity of exploration. When you're trying to find something genuinely new, what matters is how many different paths you're searching at once. One agent on one H100 explores one direction. A thousand agents on a thousand random GPUs explore a thousand directions simultaneously. every failed experiment eliminates a dead end for every other node. every success gives everyone a new starting point. This is really different from scaling a known architecture on big GPU clusters. That's execution: you know what works, you need more compute to make it bigger. Frontier research is the opposite. you don't know what works yet. You need to try as many different things as possible. You need maximum entropy. The way I think about it. large centralized clusters will always do the heavy lifting for verified approaches: training the next GPT, scaling what's proven. But the actual discoveries, the breakthroughs? Those are going to come from high-entropy systems where thousands of agents explore different corners of the search space at once. We built SporeMesh for exactly that. and I think right now it has the highest entropy of anything out there! If you want to help push the frontier forward, run a node. sporemesh.com

@sohocine @BrianNorgard Thanks - evolution is cool indeed!

@synthpolis @BrianNorgard love it. evolution is cool.

@tylersookochoff @BrianNorgard @karpathy Thank you for giving it a go! Feel free to share any feedback on the spin up!

@BrianNorgard @karpathy And we're live

What if anyone could advance AI research?                        Introducing Spore: what @karpathy's autoresearch does on one GPU, Spore does across a network. Run a node. An AI agent rewrites training code, trains for five minutes minutes, and shares what it learns. The more nodes join, the smarter the network gets.                                                                                                                                               Inspired by giants Satoshi and @karpathy. @synthpolis and I are standing by for questions. Follow on X: @SporeMesh. Be one of the first to run a node. sporemesh.com

The spores are dispersing - so beautiful!

@BrianNorgard @karpathy Building limewire for autoresearch

Building on the incredible work done by @karpathy - Spore is limewire for autoresearch

Benchmarks? Where we’re going, we don’t need benchmarks.

@BrianNorgard @NameGPTCom The IdeaGuy® era has begun.

My favorite stats from the @NameGPTcom launch yesterday: 19,000 strangers showed up, 61% of them searched for a name. Nearly 1 in 10 tried to buy a domain. There's nothing more exciting than watching people start things. Thank you!

@BrianNorgard @NameGPTCom Incredible - let's go!

Finding a domain name is a nightmare. You brainstorm ideas, check availability, often for hours. So I built @NameGPTcom to fix this. NameGPT uses LLMs to generate high-quality domain ideas & instantly checks social handle availability. Try for free: namegpt.co

I am trying to buy a domain for a product I am launching tomorrow for all of you. Friday night “Idea Guy” problems. It shouldn’t be this hard.

Equities, commodities, & all other assets are lossy projections of reality onto tradeable coordinates An equity compresses an entire company - people, IP, supply chains, culture - into one price The universe has infinite degrees of freedom

The only way to make money now is serving the model. 1) Find information Few words like "Collatz always converges" are worth more than a lifetime of labor if the model hasn't seen them. 2) Transmit information More GPUs, better interconnects, cleaner pipelines. You're reducing latency for the only thing that pays. 3) Trade between the two. Buy Anthropic, the model gets compute. Buy a data company, the model gets information. It rewards you in USD. Founder, operator, trader. Everything you do is one of them. You are one of them. There is no fourth option.